In order to predict more accurately the frequencies of lane changes with the limited data, a combined forecasting model based on neural network and Markov chain was proposed. The average speed and density of road sections were collected, and the BP neural network model was used to train the preliminary fitting model. By using Markov chain method, the distribution and probability of lane changing frequencies in three groups of intervals representing overestimation, normal and underestimation were given to decrease BP neural network er- ror. The combination forecasting model was adopted to predict and analyze the frequency of lane-changing of a road in Xi'an. The results show that the actual frequencies of lane changing are all within the prediction range of the maximum probability given by the model, which indicates that the model can predict the frequency of lane changing according to the average speed and traffic density of the section.